At Silverpush, I build and own production machine-learning pipelines for contextual advertising and content intelligence. I shipped a subtitle extraction system covering 2,137,577 unique content identifiers and built data-cleaning scripts that support 98.63% plot coverage across 1,214,713 identifiers.
I also own the Trending-Topic Intelligence Pipeline, combining clustering, classification and LLM-backed systems. I reworked its LLM stages to reduce daily API spend from roughly USD 25–30 to USD 3 with no degradation to output, and built a memory-aware GPU scheduling layer to prevent failures caused by GPU contention.

